ITR: Efficient Self-Organizing Content Distribution Network for Scalable Video Streaming Services

作为IP组播的后继者，应用层组播和覆盖网络已经成为当前Internet中非常重要的两种技术，它们承诺为向大量Internet用户并发传递信息提供高度可扩展的平台。本项目专注于下一代覆盖网络，并研究为实时视频流服务构建分布式、自组织内容分发架构的问题。与以前的方法不同，我们的工作不依赖于延迟作为距离度量，而是利用现有互联网的拓扑和路由结构来计算距离并创建本地拓扑。节点利用拓扑感知发现每个复制服务器的邻域，为静态内容的复制构建有效的网格，并为实时直播视频材料的流构建自组织多播覆盖树。需要特别注意的是确保整个体系结构具有高度可伸缩性，并且适合在现有Internet上的实际使用。安装后，CDN （content distribution network）中的每个节点自动发现自己在Internet中的位置，附着在CDN结构的控制树上，并在其附近找到其他节点。控制树中的信息包含CDN网络节点独立发现的所有Internet视图的可扩展聚合。使用此控制树，节点可以快速构建一组重叠的邻域网格，并到达主数据服务器。到达的最终用户根据其在互联网和网络集群中的位置被有效地定向到最近的缓存，这些位置已经被各个缓存节点发现。为了避免丢包和控制沿多播树分支的流，我们研究了新的视频流拥塞控制方法，使视频流能够在网络延迟反馈下平稳播放并保持稳定。cdn作为现有网络基础设施的重要组成部分出现，并且已经改善了许多互联网下载的响应时间和吞吐量。未来支持视频流的cdn将提供一个可扩展的平台，用于向最终用户分发直播和预录制的视频节目，并将大大提高当前社会传播信息的能力。建议的研究活动将整合到网络课程课程中。这将使学生深入了解与点对点网络和内容分发网络相关的建筑设计问题。

As successors of IP multicast, application-layer multicast and overlay networks have become two very important technologies in the current Internet that promise to provide a highly scalable platform for concurrent delivery of information to large groups of Internet users. This project focuses on next-generation overlay networks and studies the problem of building a distributed, self-organizing content distribution architecture for real-time video streaming services. Unlike previous approaches, our work does not rely on delay as the metric of distance, but rather utilizes the topological and routing structure of the existing Internet to compute distances and create local topologies. Using topologically aware discovery of the neighborhood of each replica server, nodes build efficient meshes for replication of static content and construct self-organizing multicast overlay trees for real-time streaming of live video material. A special attention is paid to making sure that the entire architecture is highly scalable and suitable for practical use over the existing Internet. Once installed, each node in the proposed content distribution network (CDN) automatically discovers its location in the Internet, attaches to the control tree of the CDN structure, and finds other nodes in its neighborhood. The information in the control tree contains a scalable aggregation of all independently discovered views of the Internet by the nodes of the CDN network. Using this control tree, the nodes quickly build an overlapping set of neighborhood meshes that reach the main data server. Arriving end-users are efficiently directed to the nearest cache based on their location in the Internet and network clusters already discovered by the individual cache nodes. To avoid packet loss and to control the flows along the branches of the multicast tree, we investigate new congestion control methods for video streaming that allow smooth playback and remain stable under delayed feedback from the network. CDNs emerged as a crucial part of the existing web infrastructure and already improve the response time, as well as the throughput, of many Internet downloads. Future CDNs that support video streaming will deliver a scalable platform for distributing live and pre-recorded video programs to end-users and will greatly improve the current ability of the society to disseminate information. The proposed research activities will be integrated in the network course curriculum. This will enable students to gain deep understanding of the architectural design issues related to peer-to-peer networking and content distribution networks.